The seeder inspects the query, the target page, and the competitive context to output testable causal claims alongside initial prior probabilities estimating each claim's likelihood of improving rank.